This is the github repository of The AIStats 2024 paper:
"DE-HNN: An effective neural model for Circuit Netlist representation.
Z. Luo, T. Hy, P. Tabaghi, D. Koh, M. Defferrard, E. Rezaei, R. Carey, R. Davis, R. Jain and Y. Wang. 27th Intl. Conf. Artificial Intelligence and Statistics (AISTATS) 2024." arxiv
The dependencies are recorded in requirements.txt. You can install the dependencies using pip:
pip install -r requirements.txt
A seperate README for Netlist Data is available here: Data README
Raw data is available at link.
After downloading the raw data, please extract and put the raw data directory to "/de_hnn/data/" directory.
After that, please run following command to generate the full processed dataset. The whole process can last for hours.
source run_all_python_scripts.sh
If one does not want to process the raw data for any reason but just need the processed data, the full processed data is also available at link.
After downloading the processed data, please extract and put the data directory to "/de_hnn/data/" directory.
After the data is processed, the experiments can be run following the instrcutions below.
First, enter directory "/de_hnn/experiments/cross_design/".
A simple test on cross-design full-DE-HNN can be run using the following command:
source run_simple_test.sh
A training process will be initiated immediately. This is a simple test just to test if codes are ready to use.
A more complete test run with full number of epoches can be run using the following command:
source run_full_test.sh
The whole process might take 30 minutes to hours depending on devices, which will produce the full results for full-DE-HNN for cross-design.
All files are ready in directory "/de_gnn/experiments/single_design/".
Please run the commands in this format "source run_all_{model}.sh" depending on which model you want to run.
All files are ready in directory "/de_gnn/experiments/cross_design/".
Please run the commands in this format "source run_all_{model}.sh" depending on which model you want to run.
Thank you!